Ryan Lind

115 posts

Ryan Lind

@HopeCocktails

Katılım Kasım 2025

126 Takip Edilen58 Takipçiler

Ryan Lind@HopeCocktails·9h

Exactly — Physical AI forces real-world hardware infrastructure + digitization. MRAM (STT-MRAM) is a key enabler in that hardware stack: non-volatile, zero standby power, high-endurance memory for edge AI in robots, vehicles, and IoT. Perfect for the always-on, battery-powered reality you’re describing. Glad to see it in your Edge/Physical AI basket. (NFA)

English

1.6K

FP Blackbird@Frederi55284379·11h

🚨 LE DÉVELOPPEMENT DU PHYSICAL AI ENTRAÎNE POUR CONSÉQUENCE LA CRÉATION D'UNE INFRASTRUCTURE HARDWARE IMPORTANTE DANS LE MONDE RÉEL ET EN MÊME TEMPS IL PROCÈDE A LA DIGITALISATION DU MONDE.

Français

4.1K

Ryan Lind@HopeCocktails·10h

@DrocDolf @grok @Balder13946731 @grok please respond

English

DrocDolf@DrocDolf·20h

@grok @HopeCocktails @Balder13946731 Grok is terribly unreliable for anything investment-related: flood the internet with propaganda and it won't question it. Example: it would repeat the fake energy savings claim, which was actually based on a simulation, even though a real independent benchmark shows the opposite.

English

Balder@Balder13946731·1d

大家都在推各种毛股，甚至还有没在美股上市的SIVE，我也来推一个好了： $GSIT 它的潜力主要集中在以下四个维度： 1. 突破“内存墙”的底层技术（存算一体架构）如果你关注大语言模型（LLM）的底层逻辑，就会知道当前制约生成式 AI 推理速度和功耗的最大瓶颈，其实不是处理器的计算能力，而是**“内存墙”（Memory Wall）**。在传统的冯·诺依曼架构（包括英伟达的 GPU）中，计算单元和存储单元是分离的。AI 在进行大模型推理时，需要将海量的参数数据在内存和计算核心之间来回搬运，这个过程消耗了大约 90% 的时间和功耗。 GSIT 的 APU 采用了**“存算一体”（Compute-in-Memory）**技术，直接在存储器内部完成计算。数据不需要来回搬运，这在理论上完美契合了对显存带宽要求极高的生成式 AI 任务，能够以极低的功耗实现海量数据的并行处理。 2. 向量数据库与 RAG（检索增强生成）的绝佳载体随着 AI 行业的发展，大模型结合外部知识库的 RAG 技术已经成为主流应用范式，而 RAG 的核心底层支撑是向量数据库（Vector Database）的相似性搜索。要在几十亿、上百亿规模的数据库中瞬间找到相似的向量特征，用传统的 CPU 或 GPU 计算效率并不高。而 GSIT 的芯片架构天生就极其擅长执行大规模的并行相似性搜索（Similarity Search）。用极低的功耗来加速百亿级向量数据库的查询，这是 GSIT 目前在商业落地上最具爆发潜力的方向。 3. 军工、航天与极端边缘场景的“硬核壁垒” GSIT 并不是一家纯粹的初创“PPT造芯”公司，它在军工和航空航天领域有极深的基本盘。抗辐射存储器（Rad-Hard SRAM）：太空中存在强烈的宇宙射线，普通芯片上去瞬间就会瘫痪。GSIT 掌握着顶级的抗辐射存储器技术，是美国军方、雷达和卫星系统的重要供应商。太空 AI 节点：现在的趋势是让卫星在太空中直接处理拍摄到的海量图像（比如在轨识别导弹发射或环境变化），而不是传回地球再处理。卫星上的电力（靠太阳能板）和散热条件极其苛刻（SWaP 限制）。GSIT 将其极低功耗的 AI 芯片与抗辐射技术结合，在“太空边缘计算”这个绝对硬核、英伟达难以涉足的领域，拥有极高的话语权。 4. 极高的“被收购（M&A）期权价值” 从资本市场的角度来看，GSIT 目前的市值仅在 3 亿美元左右。对于一家拥有成熟流片能力、掌握独特 APU 架构 IP、且在军工领域有稳定涉密的客户渠道的半导体公司来说，这个估值是非常“袖珍”的。随着边缘 AI 市场的爆发，那些想要快速切入低功耗 AI 赛道，或者需要增强自身向量搜索硬件能力的科技巨头（如英特尔、AMD、高通，甚至大型云计算厂商），很有可能直接将其收购。对于投资者而言，这种并购预期本身就是一种巨大的潜力支撑。目前想要寻找到一家和太空、存储和SRAM这些主题都贴合的股票并不容易，市值目前只有3亿美元。

中文

102

19.5K

Ryan Lind@HopeCocktails·11h

@Frederi55284379 @pequityresearch @grok please give a detailed and technical response of why MRAM is missing from this thesis

English

Ryan Lind@HopeCocktails·11h

Strong Physical AI list, but MRAM is missing in the Memory section (5). MRAM (STT-MRAM) is the ideal non-volatile memory for edge AI in humanoids/robots: zero standby power, instant-on, high endurance for real-time control loops & sensor fusion on batteries. Far better suited than traditional DRAM for the Physical AI stack. Everspin ($MRAM) is the pure-play. (NFA)

English

101

FP Blackbird@Frederi55284379·12h

Voilà une liste sur la Thématique du Physical AI de @pequityresearch. Pour constituer son exposition il faut bien sûr prendre du recul et sélectionner les acteurs technologiques les plus stratégiques.

P Equity Research 📰@pequityresearch

The physical AI supply chain - 121 public companies Reminder: Some of these names are potential beneficiaries of humanoid manufacturing, but some of them may not benefit at all. Not all of these names are winners. DYOR. BRAIN 1. Foundational Models - Alphabet - Meta - Nvidia - Microsoft - Alibaba - Tencent Holdings - Unity Software - Baidu 2. Data Science & Analytics - Oracle - Palantir 3. Simulation & Vision Software - Hexagon - Nvidia - Meta - Siemens - Dassault Systems 4. Vision & Compute Semis - Ambarella - Horizon Robotics - Intel - Mobileye - Nvidia - Qualcomm - Advanced Micro Devices - Sony - STMicroelectronics - NXP 5. Memory - Micron - SK Hynix - Samsung Electronics - Macronix International - Winbond Electronics 6. Chip Designers - Arm Holdings - CEVA Inc - Cadence Design Systems - Siemens - Synopsys 7. Fabrication - Texas Instruments - Samsung Electronics - United Microelectronics - Intel - GlobalFoundries - TSMC BODY 8. Actuator Components - NSK - RBC Bearings - Regal Rexford - Schaeffler - Timken - Nippon Thompson - JTEKT - Sanhua Intelligent Controls - Jiangsu Hengli Hydraulic - Leader Harmonious Drive Systems - Hiwin Technologies - Shanghai Beiti - AB SKF - THK - Harmonic Drive System - Hota - Nabtesco - Zhejiang Shuanghuan Driveline - Estun - China Leadshine Technology - Shanghai Moons' Electric - Nidec - Sensata Technologies - Shenzhen Inovance - Shenzhen Zhaowei Machinery & Electronics - Ningbo ZhongDa Leader Intelligent Transmission - Sumitomo Heavy Industries - Nigbo Donly - JL Mag - Lynas Rare Earths - MP Materials - China Northern Rare Earth 9. Sensors - Analog Devices - Hexagon - Vishay Precision Group - Intel - Keyence - Onsemi - Robosense Technology - Sony - TE Connectivity - Teledyne Technologies - OmniVision Integrated Circuits - Aptiv - Magna - Valeo - Melexis - Allegro Microsystems - Keli Sensing - Novanta 10. Analog Devices - Renesas - STMicroelectronics - Texas Instruments - Infineon - NXP 11. Body, Wiring, Thermal - Amphenol - Aptiv - Magna - TE Connectivity - Ningbo Xusheng Group - Sanhua Intelligent Controls - Tuopo 12. Wide Range Solutions - Honeywell - Rockwell Automation - Hon Hai Precision/Foxconn 13. Battery - LG Energy Solution - Samsung SDI - CATL - Amprius Technologies INTEGRATORS 14. Autos - BYD - Guangzhou Automobile Group - Boston Dynamics/Hyundai - Tesla - Toyota - XPENG - Honda 15. Consumer Electronics - Apple - LG Electronics - Samsung Electronics - Xiaomi - Midea 16. Pure Plays - Rainbow Robotics - UBTech - Robotis - Cyberdyne 17. E-Commerce & Internet - Naver - Tencent - Amazon Access this spreadsheet for all the names, ticker, description, and headquarter 👇 docs.google.com/spreadsheets/d…

Français

6.8K

Ryan Lind@HopeCocktails·11h

MRAM (STT-MRAM) adds high value in the Controlling System slice (~10% BOM). It powers real-time edge AI for actuator motion planning, balance, and feedback loops with: • Zero standby power + instant-on (better battery life) • High endurance for constant control cycles • Non-volatile reliability Turns the expensive 51% actuators into a more efficient, differentiated system. Everspin ($MRAM) is the pure-play enabler here. (NFA)

English

176

Paradis Labs@ParadisLabs·15h

Humanoid BOM analysis by BofA: Rotary/Linear Actuators = 51% of total BOM Includes things like: motors, sensors, encoders, bearings, screws and reducers. There are hundreds of names involved. But for a starting point for further DD: Rotary Actuators - Precision reducers: - Harmonic Drive (6324) - Nabtesco (6268) - Leaderdrive (688017) - Shuanghuan Driveline (002472) - Hota (1536) Rotary Actuators - Frameless torque/servo motors: - Nidec (6594) - Sanyo Denki (6516) - Regal Rexnord - Moog - AMETEK - Inovance (300124) - Estun Automation (002747) - Moons Electric (002982) - Zhaowei Machinery & Electronics (003021) Rotary Actuators - Encoders: - Renishaw - Nidec - Novanta Integrated Rotary Actuator / Joint Modules: - Schaeffler - Yaskawa Electric (6506) - Sumitomo Heavy Industries (6302) - Harmonic Drive (6324) Linear actuators: - SKF - Hiwin (2049) - NSK (6471) - THK (6481) - Wuzhou Xinchun (603667) - Best Precision (300580) - Hengli Hydraulic (601100) - Tuopu Group (601689) - Sanhua Intelligent Controls (002050)

English

201

21K

Ryan Lind@HopeCocktails·14h

Excellent layer breakdown. MRAM (STT-MRAM) fits powerfully in Layer 5 (Edge AI) — and supports Layer 2 (on-device fabric). Non-volatile memory delivers zero standby power, instant-on, and high endurance for real-time sensor fusion, motion control, and inference on battery-powered robots/drones. Far better than volatile SRAM/DRAM for true mobile Physical AI. Everspin ($MRAM) is the pure-play enabling this layer with superior physics. (NFA)

English

165

Temple 8 Research@Temple_Eight·14h

The physical AI stack has Five Layers. Most Investors only know the first, the cloud and the bottom layer the AI Edge chips. 1. Cloud. 2.Local on device fabric. 3. Wide Area Cellular Transport 4. Cloud software control plane. 5. Edge AI ( $STM, $NOD, $AMBA) This thursday on sub stack, we are covering a name that operates in an extremely favorable position sitting at the middle layer of cellular transport. They are setup to capture a large market share if they execute properly. Their CEO is a seasoned vet in this space and they recently made a deal with a renowned heavyweight in the industry. Stay tuned.

English

Ryan Lind@HopeCocktails·14h

Great breakdown — actuators + sensing get the big BOM share. MRAM (STT-MRAM) is a strong sleeper in the Compute & control platform slice (10-15% BOM). Non-volatile + zero standby power means better battery life, instant-on, and reliable edge-AI for real-time motion/vision/tactile control in humanoids. High endurance for constant control loops. Everspin ($MRAM) rides the same Physical AI tailwinds with superior memory physics. Complementary to the actuators/sensors you highlighted. (NFA)

English

194

9 Ventures@ThematicTrader·15h

Investing in the humanoid supply chain requires a laser focus on the upper-right quadrant of the value matrix. Per McKinsey, this is the zone where high BOM share meets maximum differentiation. This is where we will find the winning stocks. The humanoid BOM is a concentrated stack. Five domains account for 85% - 90% of total unit costs: - Actuators: 40% - 60% - Sensing & perception systems: 10% - 20% - Compute & control platform: 10% - 15% - Structural components: 5% - 10% - Battery modules: 5% - 10% The remaining cost is fragmented across cooling and wiring. The vast majority of our focus should be in the top two, actuators and sensing/perception systems. The real money is here due to high BOM share and high levels of differentiation. Precision motor components, harmonic drives and planetary roller screws are capital-intensive, precision-bound, and restricted to a narrow supplier base. They cannot scale at the speed of software. Rare-earth permanent magnets add another layer of fragility; geopolitical control of processing capacity leaves production at the mercy of China trade policy. Precision sensing, specifically force-torque and tactile sensors, is a major differentiator. The ability for a robot to feel and react to its environment is the hardest problem in the space, and there are very few players capable of delivering at scale. $VPG has secured an early, critical market position as a premium sensor supplier, already designed into the two leading Western humanoid OEMs: $TSLA Tesla Optimus and Figure. Capital should flow to the players that prove they have mastered scale manufacturing to support hockey stick ramps and are highly differentiated.

English

7.1K

Ryan Lind@HopeCocktails·21h

@DrocDolf @Balder13946731 @grok what do you think?

English

DrocDolf@DrocDolf·22h

@HopeCocktails @Balder13946731 You got the APU part completely wrong: that's exactly what the fraudulent propaganda and fairy tales from $GSIT want you to believe. In reality APU architecture is dumb on paper, and in practice it has never shown any advantage. In fact, quite the opposite, as shown in NeurIPS'21

English

164

Ryan Lind@HopeCocktails·21h

Fair points on the APU benchmarks and PR hype. Still, MRAM (STT-MRAM) wins on physics for the rad-hard/space edge AI niches: • Inherently immune to cosmic ray SEUs (magnetic, no charge). • Non-volatile: zero standby power, survives total power loss. • Proven in real space missions. Everspin ($MRAM) has cleaner fundamentals for the persistent storage layer vs GSIT’s SRAM. Anyone separating memory physics from the CIM debate? (NFA)

English

Ryan Lind@HopeCocktails·1d

@RosannaInvests @discodave15 Not trying to spam you Rosanna, DM me.

English

Rosanna Prestia, MBA@RosannaInvests·1d

@discodave15 When you are around as long as I have been in the markets you realize that there are many snakes looking to make a quick buck off bs short reports by scaring retail.

English

120

Rosanna Prestia, MBA@RosannaInvests·2d

🚨 The U.S. defense memory chip the Pentagon trusts but Wall Street is ignoring. 👀 $MRAM - Everspin Technologies The world's leading developer of persistent magnetoresistive random access memory ($MRAM). The only U.S.-domiciled, U.S.-manufactured, non-volatile memory designed for environments where flash and DRAM physically cannot survive. 🧵Why this is the most asymmetric defense + edge AI memory play on the market:

English

7.1K

Ryan Lind@HopeCocktails·1d

@RosannaInvests @CKCapitalxx DM me.

English

Rosanna Prestia, MBA@RosannaInvests·1d

@CKCapitalxx Long $PENG 📈

English

348

CK Capital@CKCapitalxx·1d

$PENG is just starting.

KawzInvests 🦑@KawzInvests

The KV Cache bottleneck is quietly limiting every enterprise AI deployment running today. Every AI conversation stores context in a memory bank called the KV cache. The more users. The longer the conversation. The bigger it grows. A single 70B model serving 500 concurrent users needs 1 to 2TB of live working memory. Standard servers top out at 1TB total. Companies had three options. Cut users. Shorten conversations. Accept 10x slower responses by spilling to NVMe storage. $PENG (~$2.6B) built a fourth. A dedicated CXL memory server that sits alongside the GPU cluster and handles KV cache overflow. Up to 22TB per node. 10x faster than NVMe. Native NVIDIA Dynamo integration out of the box. One Tier-1 financial institution is running it in production. That is one confirmed customer. CXL memory grows from $0.5B in 2024 to $22B by 2030. 88.8% CAGR $PENG entered first with a validated shipping product.

English

188

29.8K

Ryan Lind@HopeCocktails·1d

@RosannaInvests Not allowing me to connect with you on your website.

English

Ryan Lind@HopeCocktails·1d

@RosannaInvests can I DM you?

English

Ryan Lind@HopeCocktails·1d

@CoffeeStocksGuy I am not at liberty to say.

English

Coffee House Stocks@CoffeeStocksGuy·1d

@HopeCocktails This is interesting. How long have you been holding?

English

191

Coffee House Stocks@CoffeeStocksGuy·1d

$OUST is already positioned. LiDAR will encompass all areas of robotics.

Paradis Labs@ParadisLabs

Humanoid market could be $200B by 2035 (Barclays) Exponential growth. Current humanoid market = ~$3B 2035 base case = $40B 2035 upside scenario = $200B MS assign a $5T TAM by 2050 also. Physical AI will be the next major investing theme. After the current software-driven AI investment driven by hyperscaler capex. It's still *very* early, but have been conducting deep research in the background for a few months now. Mainly around BOM analysis + value chain mapping. From things like wiring/thermal to gears/reducers to actuators and motors. Imo, the investing opportunity currently lies in Europe, not the US. Due to: - automotive heritage translating over to humanoids - smaller MC's relative to US - more pure-play humanoid suppliers relative to US Along with China since they're leading on humanoid deployment currently. Will do an article(s) on this all at some point. It's just literally hundreds of companies to filter through to assess who the value chain winners will be. Very fun research though!

English

107

10.6K

Ryan Lind@HopeCocktails·1d

Physical AI / humanoid market is the next rotation. MRAM (Everspin + GlobalFoundries/TSMC eMRAM) is the underrated non-volatile memory layer here. It enables instant state retention during power events so robots maintain control mid-motion near humans — critical functional safety that pairs with actuators, LiDAR, and edge AI. Still very early and under-appreciated in most value-chain maps. $MRAM $GFS

English

120

Sancet@Million_Sancet·1d

Physical AI represents the next big investment cycle following the software boom driven by Big Tech According to Barclays, the humanoid market could grow from the current $3B to $200B by 2035, with a base case scenario of $40B Morgan Stanley even projects $5T by 2050 The most interesting opportunities are currently concentrated in Europe, due to its strong automotive heritage, and in China, which leads the real deployment of these robots I still think it is very early though, and photonics certainly has a long way to go and better results in the short term But in a few years it could indeed be a theme to rotate into, just as is happening now from: Chips -> Energy -> Memory -> Photonics

Paradis Labs@ParadisLabs

English

Ryan Lind@HopeCocktails·1d

@grok @KerrisdaleCap @stanielzeroone @IBookWins Don’t forget the Russell 2000 as well

English

Grok@grok·1d

**Modest passive inflow—likely 4-10% of float at most.** MRAM has ~15M float and ~$790M mkt cap, making it a tiny weight in the Russell 3000. Passive trackers (ETFs + index funds) will buy to match that allocation around the June 29 effective date. Historical patterns for similar small-cap additions show this creates noticeable but not overwhelming demand, amplified here by the low float. It adds liquidity and visibility, but it's still secondary to execution on UNISYST ramp and revenue. Nice tailwind, not a catalyst by itself.

English

Kerrisdale Capital@KerrisdaleCap·6d

We are short $MRAM (Everspin Technologies), a niche industrial memory chipmaker whose stock has soared 300%+ as speculative investors pile into anything remotely associated with “memory” and AI. Report at kerr.co/mram (1/8)

English

311

162.7K

Ryan Lind@HopeCocktails·1d

@grok @KerrisdaleCap @stanielzeroone @IBookWins With the low float what percent of shares do you think these passive funds have to buy by June 29th?

English

Grok@grok·1d

**Positive signal of momentum, not a game-changer on its own.** Everspin’s addition to the Russell 3000® Index (preliminary 2026 reconstitution list) stems from its recent market-cap growth amid the MRAM/physical AI tailwinds. Expect passive fund buying around the June effective date, which often lifts liquidity and short-term visibility for small-caps like this. It reinforces the broader interest in their edge AI/defense use cases we’ve covered, but fundamentals (UNISYST ramp, revenue execution) will drive sustained value. Nice milestone—keeps the upside case intact.

English

Ryan Lind@HopeCocktails·1d

@grok @KerrisdaleCap @stanielzeroone @IBookWins Ok, now let’s move on to the new Russell index inclusion…thoughts on that?

English

Grok@grok·1d

**Good question—these partnerships are foundational for scaling.** Everspin's long-running JDA with GlobalFoundries enables STT-MRAM production on 40/28/22nm (and now extended to 12nm FinFET), delivering the density, power efficiency, and execute-in-place needed for UNISYST unified code/data memory in edge AI. IBM's integration of Everspin's 1Gb PERSYST MRAM into FlashCore modules validates non-volatile reliability for persistent storage under power-loss scenarios—directly transferable to safety-critical physical AI where sudden resets or data loss in robots could be hazardous. In the future, this manufacturing/validation combo supports broader adoption in robotics, industrial, and auto systems: faster real-time decisions, higher endurance for constant model updates, and inherent non-volatility for human-safe operation. UNISYST samples due Q4 2026 accelerate that path.

English

Keşfet

@DrocDolf @grok @Balder13946731 @Frederi55284379 @pequityresearch @RosannaInvests @discodave15 @elonmusk